Cigna's PxDx post-service claim review

What this models

This example runs on the PxDx-class post-service batch claim review network: 11 components and 25 pathways between them. Every context in the Lab is a stylized model, never a reconstruction of any actual deployment, and each assumption behind it carries a provenance label.

Evidence base: 1 assumed · 9 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

• baseline

  D48-derived new org (Phase 6, clinical-decision-support). TOPOLOGY. Eleven nodes, all documented, none decorative - and the POST-SERVICE position is the point: every other org in this domain adjudicates before or during care, while this one decides after the clinical decision is already unchangeable, so what it allocates is payment and appeal burden rather than access. One model, because the record documents one deterministic screen and both sides of the dispute agree there is no machine learning in it. Three stores, because the record documents three with three different readers: the payer-authored match list the screen compares against, the working determination record, and the review policy FILED with the state regulator, which the Department's October 2025 findings say differed from the process in use. A worklist, because the queue of flagged mismatches worked in batches is documented as a queue and is the artifact the case turns on. An enforcement node, because a signed mismatch issues as a denial letter and a payment refusal as a matter of course. Two reviewers, because the correction path is documented in two distinct stages with different positions - a second Cigna doctor inside the plan, and an independent review organisation outside it. FOUR ABSENCES ARE EQUALLY DERIVED: no operatorToModel edge of any kind (the screen takes no per-case operator input at all - no prompt, no completion, no confirmation loop - and the human contribution to what it decides is authorship of the list, drawn as a store write); no externalBoundary (the system is in-house, there is no vendor, no data broker and no documented replication out of the governed system); no guardrail (no automated screen sits on the model's output - the documented pre-issuance control is a human signature); and no second model (there is nothing to cross-check with, and the model self-check pathway is drawn at zero for that documented reason).

• baseline

  DEMAND 3 / CAPACITY 3, and the two are answering different questions. Demand 3: the documented load is over 300,000 payment requests denied through this method across a two-month window in 2022 at an average of 1.2 seconds of physician attention per case, with individual directors signing 60,000 to 121,000 denials in one to two months - the ProPublica and Capitol Forum computation from internal company records, which Cigna disputes without publishing substitute figures. Capacity 3 is read against the field's own definition, what the human process delivers with no AI at all: here that counterfactual is a licensed physician reading the individual claim's clinical context, which is what California Health and Safety Code 1367.01(e) requires for a medical-necessity denial, what the plan's own filed policy described, and what the external review organisation did in the one arc the record follows to the end. This is the MiDAS reading applied honestly - automation placed over a review the law requires a person to perform, so this network must be able to read net-negative. The thinness of the CURRENT read is deliberately not folded in here; it is carried on the near-floor rung of the pathway from the claim record into the signature layer, which is where it belongs.

• baseline

  BASELINES mirror the PAN org's evidence-derived edge widths as intensity rungs (the full mapping, with every deviation and its reason, is in the derivation notes). Three structural contrasts are deliberate and documented. First, the match list runs into the screen at 3 while what the signature layer reads out of the claim record runs at 1: the decision boundary is wide and the read at the point of determination is at the floor, which is the ProPublica and Capitol Forum finding and the California regulator's finding drawn as a width asymmetry rather than asserted in prose. Second, the outside correction's per-item effect runs at 2 while the pathway feeding it runs at 1 - the channel reaches the right answer on what gets to it, and the investigation reports the plan's own expectation that about 5 percent of people would appeal at all, on claims selected for being low-dollar. Averaging that pair into one middling number is exactly what this network refuses. Third, the two appeal pathways BOTH run at 1 even though the internal channel is documented as wider than the external one, because the same design expectation gates both and the intensity vocabulary has no rung between absent and low; the ordering is carried in the attached copy rather than in a rung that cannot honestly express it.

• baseline

  THE SAVINGS LOOP is this network's signature pathway and the reason its store side is drawn at all. Additions to the match list are justified by projected savings from denying the added category, and those savings are computed from the denials the list produces, so the boundary grows on its own output. The investigation documents the mechanism from internal company records with one dated instance - autonomic-nervous-system testing added in 2014 on a projection of roughly 2.4 million dollars a year - and quotes the executive credited with building the process saying it had undoubtedly saved billions. It runs at 2 rather than 3 because the record establishes the mechanism and one instance rather than a rate, and because list additions are episodic authorship rather than a continuous replication. Cigna disputes the article's characterization of the process. No sibling in this domain has a decision boundary whose growth is funded by what it denies.

• baseline

  THE FILED-POLICY STORE, and why an oversight channel is drawn reading a different record than the one running. Everything the state managed-care regulator formally knew about this review process arrived through the policy the plan filed with it. On 8 October 2025 the Department of Managed Health Care fined Cigna HealthCare of California 500,000 dollars, finding that the plan reviewed and denied claims without physicians conducting clinical reviews of the claims prior to issuing denials, and that it used a different review process than the policy it had filed with the Department; the agreed corrective actions include re-reviewing denials issued under the non-compliant process going back two years and revising and refiling the policy. Those are regulator findings agreed to by the regulated entity and are stated here as such. THE LINKAGE CAVEAT IS LOAD-BEARING: the Department's release names neither PxDx nor the investigation, so this network treats the action as addressing the plan's claims-review practice, consistent with the documented pattern, and never as a finding about PxDx by name. The respondent is the California-regulated plan entity, not the national group.

• baseline

  THE RETROACTIVE CHECK. The reconciliation drawn back from the denial letter onto the claim record is present at 1, not 0, because the obligation is real, dated and agreed: the October 2025 settlement requires this plan to re-review denials issued under the non-compliant process going back two years. It is the rarest thing in this cluster - a correction that reaches determinations nobody appealed - and it is drawn at the low rung rather than higher for three documented reasons: it binds one state's regulated entity rather than the national book, it was agreed once rather than established as a standing reconciliation, and no report of its completion or its results has been located. Nothing here asserts that the re-review has been carried out or what it found.

• baseline

  WHAT ERROR MEANS HERE, AND WHY NO ACCURACY NUMBER APPEARS. The system is deterministic, so an error on this channel is not a mis-estimate of anything: it is a flagged mismatch that a clinical review of the individual claim would not have denied under the governing coverage terms. That makes the error entirely a property of the payer-authored list and of what a flag is permitted to become downstream. No deployment error rate is published, no independent evaluation of this deployment exists in the record read here, and the values on this diagram are ORDERED rather than set by two findings and one arc: the regulator's October 2025 finding that denials issued without prior physician clinical review (a process finding, not a rate), the one documented external-review reversal, and the operator's own statement that 94 percent of claims subject to the review are automatically approved and paid. The appeal and overturn statistics quoted in the congressional correspondence are Medicare Advantage PRIOR-AUTHORIZATION figures used as an analogy; they measure a different programme and no value here rests on them.

• baseline

  EVIDENCE STATUS, AND WHAT THIS NETWORK REFUSES TO CARRY. The quantitative core of this record is ONE investigative computation from internal company records by two newsrooms (an April 2023 Sidney Award), which the operator publicly disputes as riddled with factual errors without publishing substitute figures, and which is now being tested in discovery with roughly 2.1 million pages produced. Every throughput figure on this diagram is attributed to that investigation wherever it appears and is never stated as adjudicated fact. Four refusals are load-bearing. (1) The 5 percent figure is the plan's internal EXPECTATION of how many people would appeal, as reported from company records - a design assumption, not a measured appeal rate - and it is drawn as pathway width into the appeal classes, never as a rate on them. (2) The pleading-stage observation that reading the plan as allowing an algorithm to decide so long as a medical director pushes the button would conflict with the plan's plain language is an interpretation ruling on allegations assumed true, used here to order one correction value downward and asserted nowhere as a finding of fact. (3) The named medical directors whose per-director volumes appear in the internal records are private individuals and are not named anywhere in this network; their volumes are carried as a range, attributed. (4) No public findings, hearing record or released production from the May 2023 congressional document inquiry has been located, and nothing here implies the inquiry concluded or found anything.

• baseline

  THE OPERATOR'S COUNTER-FRAME IS PARTIALLY LOAD-BEARING AND IS CARRIED. Cigna's own published description states that this review applies to roughly 50 common, relatively low-cost tests and procedures; that 94 percent of claims subject to it are automatically approved and paid; that denials through it are less than 1 percent of total claim volume; that it occurs after the patient has received treatment and once their physician bills for it; and that it does not involve algorithms, artificial intelligence or machine learning. That is vendor-tier evidence and it is labelled as the operator's own account - but it is corroborated in outline by the investigation, and two parts of it are structural facts this network is built on: no care is gated by anything drawn here, and the screen is a deterministic list lookup rather than a learned model. The wrong the record supports concerns the emptiness of the review layer above the match and the authorship of the list, not a model erring. The plan's declaration evidence in the litigation also established that not every denial of this insurer's runs through this review - which is why the network is scoped to the claims the list touches.

• assumed

  Served patients and providers are not in the dynamics. No coverage decision, medical-necessity judgement, clinical outcome or financial outcome for any person is computed from anything drawn here, and no score over a person is authored anywhere in this network. The throughput and volume figures on this diagram are properties of a claim stream, recorded and attributed to the investigation that computed them, never derived. The record contains no denial, appeal or overturn measurement for this review disaggregated by any characteristic of a served person, and this network does not manufacture one; the deprivation the record documents is payment for care already delivered, plus the burden of appealing it, and both live in the case file rather than on the diagram.

What this example does not show

• The quantitative core of this record is ONE investigative computation from internal company records - over 300,000 payment requests denied through this method across two months of 2022 at an average of 1.2 seconds each, per-director volumes of 60,000 to 121,000, and the roughly 2.4 million dollar projection behind one 2014 list addition. Cigna publicly disputes the article's characterization as riddled with factual errors and has published no substitute figures, and the underlying documents are now in discovery with roughly 2.1 million pages produced. Those numbers are attributed wherever they appear here and are never stated as adjudicated fact.
• The operator's counter-frame is partially load-bearing and is carried rather than rebutted: this review is post-service, so no care is gated by it; it applies to roughly 50 common, relatively low-cost tests and procedures; 94 percent of the claims subject to it are automatically approved and paid; denials through it are under 1 percent of total claim volume; and both sides agree it involves no algorithms in the machine-learning sense, no artificial intelligence and no learned model. The wrong this record supports concerns the emptiness of the review layer above a deterministic match and the authorship of the list, not a model erring.
• The California Department of Managed Health Care's October 2025 action IS a set of regulator findings agreed to by the regulated entity and is stated as such: a 500,000 dollar fine against Cigna HealthCare of California, findings that claims were reviewed and denied without physicians conducting clinical reviews of the claims prior to issuing denials and that the plan used a different review process than the policy it had filed, and agreed corrective actions including a two-year re-review of denials and a policy refiling. The Department's release names neither PxDx nor the investigation, so this scenario describes the action as addressing the plan's claims-review practice, consistent with the documented pattern, and never as a PxDx fine. The respondent is the California-regulated plan entity, not the national group, and no report of the re-review's completion or its results has been located.
• The federal litigation posture is mixed and is carried precisely. In Kisting-Leung v. Cigna Corp. (E.D. Cal., No. 2:23-cv-01477-DAD-CSK), the March 31, 2025 order dismissed the ERISA denial-of-benefits claim with leave to amend and it was never repled; the ERISA fiduciary-duty claim and the California unfair-competition claim proceed, for two remaining plaintiffs as of August 2026 after one was voluntarily dismissed on August 13, 2026; and three of the original six plaintiffs were dismissed for lack of standing after the company's declaration evidence showed their denials carried no such letters, which also establishes that not every denial of this insurer's runs through this review. Class certification is unbriefed, due from October 29, 2026, so no class exists. Everything pleaded is allegation, and the court's observation that reading the plan as allowing an algorithm to decide so long as a medical director pushes the button would conflict with its plain language is a pleading-stage interpretation ruling, not a finding of fact.
• The 5 percent figure is the company's internal EXPECTATION of how many people would appeal a denial, as reported from internal records - a design assumption, not a measured appeal rate - and it is drawn here as the width of the pathways into the appeal channels, never as a rate on them. The appeal and overturn statistics quoted in the May 2023 congressional correspondence are Medicare Advantage prior-authorization figures used as an analogy for a different programme; no value here rests on them, and no public findings, hearing record or released production from that inquiry has been located.
• Served patients and providers are not modeled. Coverage outcomes, medical-necessity judgements, clinical outcomes and out-of-pocket consequences are boundary quantities recorded in the case file; the Lab models institutional propagation through the operator network and computes no patient outcome from anything on this diagram. No denial, appeal or overturn measurement for this review disaggregated by any characteristic of a served person exists in the record, and this scenario does not manufacture one. The individual physicians whose per-director volumes appear in the internal records are private individuals and are not named.

Sources and evidence

What this example rests on, claim by claim. Every entry resolves to the same ledger the Evidence Registry publishes.

• ProPublica and The Capitol Forum (Patrick Rucker, Maya Miller, David Armstrong), computing from internal Cigna records and interviews with former employees, reported on March 25, 2023 that Cigna's PxDx review flags claims where the billed procedure code does not pair with the diagnosis code on a payer-authored list, and that company medical directors then sign the flagged denials in batches without opening patient files: over 300,000 payment requests denied through this method across a two-month period in 2022, at an average of 1.2 seconds of physician attention per case, with individual medical directors signing between roughly 60,000 and 121,000 denials in one to two months. A former Cigna doctor told the reporters, 'We literally click and submit. It takes all of 10 seconds to do 50 at a time.' The same records, per the investigation, show the match list being extended on cost grounds: adding autonomic-nervous-system testing in 2014 carried an internal projection of roughly 2.4 million dollars a year in savings, and the executive credited with developing the process said it had 'undoubtedly saved billions of dollars.' Cigna publicly disputes the article's characterization of the process - a spokesperson called a complaint built on it 'based on an article riddled with factual errors and misinformation' - and has published no substitute figures; the underlying documents are now in discovery. The investigation received the April 2023 Sidney Award. These figures are the investigation's computation and are not adjudicated fact.

  empirical
  • Investigative Rucker, P., Miller, M., & Armstrong, D. (2023, March 25). How Cigna Saves Millions by Having Its Doctors Reject Claims Without Reading Them. ProPublica, co-published with The Capitol Forum https://www.propublica.org/article/cigna-pxdx-medical-health-insurance-rejection-claims
  • Reference Sidney Hillman Foundation (2023, April). The Capitol Forum, ProPublica win April Sidney for Exposing Cigna's Policy of Rejecting Health Insurance Claims Without Reading Them https://hillmanfoundation.org/sidney-awards/capitol-forum-propublica-win-april-sidney-exposing-cignas-policy-rejecting-health
  • Trade press Healthcare Dive (2023). Cigna sued over algorithm allegedly used to deny claims; and Cigna hit with second class action suit over claims automation software https://www.healthcaredive.com/news/cigna-lawsuit-algorithm-claims-denials-california/688857/

• Cigna's own published account of PxDx, which is vendor-tier evidence and is corroborated in outline by the investigation, states that the review is 'procedure to diagnosis' code matching applied to roughly 50 common, relatively low-cost tests and procedures; that 94 percent of the claims subject to it are automatically approved and paid; that denials issued through it are 'less than 1 percent of our total volume of claims'; that the review 'occurs after the patient has received treatment and once their physician bills for the treatment'; that it 'does not involve algorithms, artificial intelligence, or machine learning'; and that in-network patients should not be billed for services denied this way. Two elements of that account are structural facts rather than contested framing: no care is gated by this review, because it runs after treatment has been delivered, so the decision allocates payment rather than access; and the screen is a deterministic list lookup rather than a learned system, which both sides of the dispute agree on. The wrong the record supports concerns the emptiness of the physician-review layer above the match and the authorship of the match list, not a model erring.

  empirical
  • Vendor The Cigna Group Newsroom. PxDx (standing explainer page); and Cigna Newsroom (2023, July 27). Cigna Healthcare Affirms its Approach to Expediting Physician Payments https://newsroom.thecignagroup.com/pxdx
  • Investigative Rucker, P., Miller, M., & Armstrong, D. (2023, March 25). How Cigna Saves Millions by Having Its Doctors Reject Claims Without Reading Them. ProPublica, co-published with The Capitol Forum https://www.propublica.org/article/cigna-pxdx-medical-health-insurance-rejection-claims

• The correction channel in this record is documented in two stages with different positions, and its reach and its per-item effect are separate facts. A denial appealed inside the plan goes to a different Cigna doctor; beyond that, an independent review organisation outside the plan can be reached. In the one patient arc the record follows end to end - a roughly 350 dollar vitamin-D blood test denied in autumn 2021 as not medically necessary - the internal appeal upheld the denial and the external independent reviewer reversed it roughly seven months after the denial was issued. What gates the channel is a design expectation rather than a measured rate: ProPublica reports from company records that Cigna internally estimated only about 5 percent of people would appeal a denial, on a set of claims selected for being low-dollar. That figure is an internal expectation about appeal propensity and must not be read as an observed appeal rate. Cigna has published no internal overturn rate for this review. The appeal and overturn statistics quoted in the May 2023 congressional correspondence (roughly one in five denials appealed, about 80 percent of appeals overturned) are Medicare Advantage prior-authorization figures the committee used as an analogy; they measure a different programme and are not measurements of this review.

  empirical
  • Investigative Rucker, P., Miller, M., & Armstrong, D. (2023, March 25). How Cigna Saves Millions by Having Its Doctors Reject Claims Without Reading Them. ProPublica, co-published with The Capitol Forum https://www.propublica.org/article/cigna-pxdx-medical-health-insurance-rejection-claims
  • Investigative Rucker, P., Miller, M., & Armstrong, D. (2023, May 16). Congressional Committee, Regulators Question Cigna System That Lets Its Doctors Deny Claims Without Reading Patient Files. ProPublica / The Capitol Forum https://www.propublica.org/article/cigna-health-insurance-denials-pxdx-congress-investigation
  • Government U.S. House Committee on Energy and Commerce (2023, May 16). E&C Republicans Press Cigna for Clarification After Investigative Report Accuses Insurance Company of Denying Claims Without Reading Them https://energycommerce.house.gov/posts/e-and-c-republicans-press-cigna-for-clarification-after-investigative-report-accuses-insurance-company-of-denying-claims-without-reading-them

• On October 8, 2025 the California Department of Managed Health Care fined Cigna HealthCare of California, Inc. 500,000 dollars for improperly denying providers' claims as not medically necessary. The Department found that the plan 'reviewed and denied claims without physicians conducting clinical reviews of the claims prior to issuing denials' and that it used a different review process than the policy it had filed with the Department. Cigna agreed to pay the fine and to corrective actions including re-reviewing denials issued under the non-compliant process going back two years and revising and refiling its review policy. Department Director Mary Watanabe said the stability of the health care delivery system is impacted when health plans wrongly deny the payment of claims for health care services. These are regulator findings agreed to by the regulated entity and are stated as fact. The scope caveat is load-bearing: the Department's release names neither PxDx nor the investigation, so this action is described as addressing the plan's claims-review practice, consistent with the documented pattern, and never as a PxDx fine; the respondent is the California-regulated plan entity rather than the national group. No report of the two-year re-review's completion or its results has been located as of August 2026. The statutory standard the parallel state-law claim rests on is California Health and Safety Code section 1367.01(e), under which no individual other than a licensed physician or a licensed health care professional competent to evaluate the specific clinical issues may deny or modify requests for authorization for reasons of medical necessity.

  empirical
  • Government California Department of Managed Health Care (2025, October 8). DMHC Fines Cigna HealthCare of California $500,000 for Improperly Denying Health Care Claims (press release) https://www.dmhc.ca.gov/Resources/Newsroom/PressReleases/October8,2025.aspx
  • Government Kisting-Leung, et al. v. Cigna Corporation, et al., No. 2:23-cv-01477-DAD-CSK (E.D. Cal.), Order Granting in Part and Denying in Part Defendants' Motion to Dismiss, Doc. 55, 31 March 2025 (Drozd, J.) https://litigationtracker.law.georgetown.edu/wp-content/uploads/2023/08/Kisting-Leung-et-al_2025.03.31_ORDER-GRANTING-IN-PART-AND-DENYING-IN-PART-DEFENDANTS-MOTION-TO-DISMISS.pdf

• The federal class litigation is live, mixed, and unadjudicated on the merits. Kisting-Leung v. Cigna Corp. (E.D. Cal., No. 2:23-cv-01477-DAD-CSK) was filed July 24, 2023. On March 31, 2025 Judge Dale A. Drozd granted in part and denied in part the motion to dismiss the third amended complaint: the ERISA section 1132(a)(1)(B) denial-of-benefits claim was dismissed with leave to amend, and plaintiffs elected on April 11, 2025 not to replead it; the ERISA section 1132(a)(3) fiduciary-duty claim proceeds; and the California unfair-competition claim proceeds on the licensed-physician-review theory. Three of the six original plaintiffs, including the named lead plaintiff, were dismissed for lack of standing after Cigna's Rule 12(b)(1) factual attack, supported by a declaration stating there were no PxDx denial letters associated with their claims - which establishes from the defense's own evidence that not every denial by this insurer runs through this review. Assuming arguendo the most deferential standard, the court held that reading the plan term requiring medical-necessity determinations by a medical director 'as allowing an algorithm to make the decision so long as a medical director pushes the button' would conflict with the plain language of the plan and constitute an abuse of discretion. That is a pleading-stage interpretation ruling on allegations assumed true, not a factual finding about what the company did. Cigna answered May 2, 2025; one further plaintiff was voluntarily dismissed by stipulation on August 13, 2026, leaving two. Per the parties' August 24, 2026 stipulation, Cigna has produced roughly 2.1 million pages and depositions are coordinated with Snyder v. The Cigna Group (D. Conn., No. 3:23-cv-1451-OAW, filed November 2, 2023), described there as another class action involving Cigna's PxDx process, so Cigna witnesses sit once for both actions; fact discovery closes in autumn 2026 and class-certification briefing begins October 29, 2026. No class has been certified and the trial setting will move. The House Energy and Commerce Committee's letter of May 16, 2023 demanded PXDX process documents, legality memoranda, the list of plans subject to the review, per-medical-director denial records, and 2022 review, denial, appeal, and overturn counts by May 30, 2023; no public committee findings, hearing record, or released production has been located as of August 2026, and nothing here implies the inquiry concluded or found anything.

  empirical
  • Government Kisting-Leung, et al. v. Cigna Corporation, et al., No. 2:23-cv-01477-DAD-CSK (E.D. Cal.), Order Granting in Part and Denying in Part Defendants' Motion to Dismiss, Doc. 55, 31 March 2025 (Drozd, J.) https://litigationtracker.law.georgetown.edu/wp-content/uploads/2023/08/Kisting-Leung-et-al_2025.03.31_ORDER-GRANTING-IN-PART-AND-DENYING-IN-PART-DEFENDANTS-MOTION-TO-DISMISS.pdf
  • Government Kisting-Leung v. Cigna Corp., Joint Stipulation and Proposed Order to Extend Deadlines, Doc. 97, 24 August 2026 https://litigationtracker.law.georgetown.edu/wp-content/uploads/2023/08/Kisting-Leung-et-al-v.-Cigna-Corp_2026.08.24_JOINT-STIPULATION.pdf
  • Reference Georgetown Law, O'Neill Institute Health Care Litigation Tracker. Kisting-Leung et al. v. Cigna Corporation et al. (tracker entry, read August 2026) https://litigationtracker.law.georgetown.edu/litigation/kisting-leung-et-al-v-cigna-corporation-et-al/
  • Government U.S. House Committee on Energy and Commerce (2023, May 16). E&C Republicans Press Cigna for Clarification After Investigative Report Accuses Insurance Company of Denying Claims Without Reading Them https://energycommerce.house.gov/posts/e-and-c-republicans-press-cigna-for-clarification-after-investigative-report-accuses-insurance-company-of-denying-claims-without-reading-them
  • Trade press Healthcare Dive (2023). Cigna sued over algorithm allegedly used to deny claims; and Cigna hit with second class action suit over claims automation software https://www.healthcaredive.com/news/cigna-lawsuit-algorithm-claims-denials-california/688857/

Where this connects

Institutional pressures in this domain

• Workload surge — Demand outruns staffing; per-case attention shrinks and review becomes triage.
• Reviewer bottleneck — One fixed-capacity checking stage sits between AI output and consequence; everything queues behind it.
• Vendor opacity — The deploying institution cannot inspect the model, data, or update pipeline it is accountable for.
• Data & policy drift — The world, the intake process, and the rules change under a system trained on how things used to be — two mechanisms with different remedies: the statistical properties of what the system processes move (concept drift), or the mixture of inputs arriving in deployment differs from the mixture it was trained on (covariate shift).
• Deadline pressure — Statutory or managerial timeliness rules reward fast approval of machine output over slow disagreement.

All of them in context on the Clinical decision support & deterioration alerting domain page.

Levers available here and the patterns behind them

• Upgrade model — Improve the model
• Mark AI-written records — Provenance labeling
• Gate record entries — Human-in-the-loop write gating
• Pause AI on alarms — Deployment circuit-breaker
• Escalate checks — State-feedback vigilance
• Assign a challenger — Structured dissent
• Understand the system — Understand the system
• Check copied records — Reconcile copied records
• Review on schedule — Oversight cadence & retrospectives
• Check with a second model — Cross-model verification

Documented case histories

• Cigna PxDx
• TREWS sepsis early-warning system
• Advance Alert Monitor (AAM) deterioration model
• Sepsis Watch deep-learning detection system
• Proprietary EHR sepsis model (external validation)
• nH Predict Utilization Review
• Cost-Proxy Care Stratification
• CA-CDS Child Abuse Alerting
• IDx-DR Autonomous Screening
• Viz.ai LVO Stroke Triage
• IBM Watson for Oncology
• OPTN eGFR Waiting-Time Correction
• Practice Fusion Pain CDS
• UBH Level of Care Guidelines (Wit v. UBH)
• EviCore by Evernorth: the review threshold